Handling shadow DOM in the various tree kinds
Flattened tree
Flattened tree is defined in CSS Shadow Module Level 1.
Shadow host
It is an Element node which has a ShadowRoot attached to it. In the
flattened tree, its children are treated as replaced with the children of the
ShadowRoot. So, the children of ShadowRoot look like the children of the
host in the flattened tree
(spec).
nsINode APIs, such as nsINode::GetChildAt<TreeKind>(),
nsINode::GetChildCount<TreeKind>() and nsINode::ComputeIndexOf<TreeKind>(),
treat the children of ShadowRoot as the children of host if TreeKind is not
TreeKind::DOM.
ShadowRoot
It is a content node, inherits DocumentFragment. Therefore, when you print
inclusive ancestors of a node in a ShadowRoot with ToString(*node).c_str(),
you’ll see #document-fragment as the root node
(spec).
In the flattened tree, the children of a ShadowRoot are formed as children
of the shadow host and the children of the shadow host are not
part of the flattened tree unless they are assigned to <slot> elements.
When comparing attached ShadowRoot and a child of the shadow host element,
ShadowRoot is treated as before the first child of the host.
HTMLSlotElement
When it appears in a shadow DOM, it may have assigned nodes which are children
of the host element. In the flattened tree, the assigned nodes of a <slot>
element are formed as its children and the children of the <slot> are not
part of the flattened tree
(spec). However, if a
<slot> does not have assigned node, the element is treated as a usual
container element because they are the fallback content of the element when
there are no assigned nodes.
nsINode::GetParentOrShadowHostNode<TreeKind>() for
TreeKind::FlatForSelection and TreeKind::Flat of a <slot> element which
has some assigned nodes return the <slot> element as the parent node.
However, with the other TreeKinds, it returns the parent node in the shadow
including DOM.
Similarly, nsINode::GetChildAt<TreeKind>(),
nsINode::GetChildCount<TreeKind>() and nsINode::ComputeIndexOf<TreeKind>()
for TreeKind::FlatForSelection and TreeKind::Flat of a <slot> which has
some assigned nodes treat the assigned nodes as children of the <slot>.
Note that the assigned nodes are not treated as in the shadow tree by
nsINode::IsInShadowTree(). So, it returns false for the assigned nodes if
they are not a descendant of a ShadowRoot.
UA shadow tree
Gecko attaches internally created ShadowRoot to some specific elements.
For example, <details>, <video> and SVG <use>. You can check it with
ShadowRoot::IsUAShadowRootSlow() (returning true means it’s a UA shadow’s
ShadowRoot). Often the children of the UA shadow host element will be assigned
to one or more <slot>s in the shadow. Therefore, they can be selected by the
user. Additionally, Selection API can
specify any points in the host element as a range boundary. However, from the
shadow including DOM point of view, children should be treated as replaced by
the children of the UA ShadowRoot. Therefore, when we handle selection
including DOM ranges, we need to ignore the UA shadow.
Unassigned node
We refer to children of a shadow host element that are not assigned to any
<slot> element in the shadow as “unassigned node”. (As far as we know, there
is no formal term for these nodes.)
Fallback content
We refer to children of <slot> that has some assigned nodes as
“fallback content” or “fallback node”. (As far as we know, there is no formal
term for these nodes.)
Unused fallback content
We refer to fallback content of a <slot> element that
has some assigned nodes as “unused fallback content”. (As far as we know, there
is no formal term for these nodes.)
Non-flattened node
We refer to inclusive descendants of an unassigned node and unused fallback content as “non-flattened node”. (As far as we know, there is no formal term for such node.)
If a node is a non-flattened node,
nsINode::GetFlatTreeAncestorElementForNonFlatTreeNode() and
nsINode::GetClosestFlatTreeAncestorElementForNonFlatTreeNode() return the
element which is an ancestor of this node and has a non-flattened child node
which is an inclusive ancestor of the node. I.e., the result is the nearest
ancestor element formed in the flattened tree. (The latter returns maybe
non-flattened node if the node is a non-flattened node. Therefore, it’s faster
than the former if you need to check whether the node is a non-flattened node.)
Kinds of DOM tree
To support the shadow DOM, there are 4 kinds of DOM trees which are identified
with TreeKind defined in nsINode.h.
TreeKind::DOM
The simplest DOM tree which do not treat ShadowRoot attached to an element.
nsINode::GetChildAt<TreeKind::DOM>(),
nsINode::GetChildCount<TreeKind::DOM>() and
nsINode::ComputeIndexOf<TreeKind::DOM>() treat the children of any nodes
as-is.
nsINode::GetParentNode<TreeKind::DOM>() returns the parent node
as-is. If you call it of a ShadowRoot, it returns nullptr.
TreeKind::ShadowIncludingDOM
This is not standardized. However, there are some points in the spec mentioned about “shadow-including tree”.
A shadow host element treats the attached ShadowRoot as
connected to its host (before the first child). However, <slot> elements are
not handled, they are always treated as usual container element.
A notable thing about this TreeKind is, the children of the host element
are also treated as its children as-is.
nsINode::GetChildAt<TreeKind::ShadowIncludingDOM>(),
nsINode::GetChildCount<TreeKind::ShadowIncludingDOM> and
nsINode::ComputeIndexOf<TreeKind::ShadowIncludingDOM>() are not available
because it’s unclear that whether they will handle the children of the host
or the ShadowRoot. Therefore, you need to use TreeKind::DOM APIs of
a shadow host or its ShadowRoot explicitly.
nsINode::GetParentNode<TreeKind::ShadowIncludingDOM>() is not available
because it’s unclear that whether it will return ShadowRoot or its host
when the node is a child of ShadowRoot. Therefore, you need to use
TreeKind::DOM API of the child or if you need to get host as a parent of a
ShadowRoot, you can use nsINode::GetParentOrShadowHostNode().
Note that the UA shadow trees are ignored when reaching the shadow DOM boundary. Therefore, the UA shadow host element is treated as same as not hosting the shadow.
TreeKind::Flat
Handle the flattened tree including the
UA shadow trees. The assigned nodes of <slot> elements are
treated as children of the <slot> element (if a <slot> does not have
assigned node, it’s treated as a normal element which may have some children).
nsINode::GetChildAt<TreeKind::Flat>(),
nsINode::GetChildCount<TreeKind::Flat>() and
nsINode::ComputeIndexOf<TreeKind::Flat>() of a <slot> which has some
assigned nodes treat the assigned nodes as the children of the <slot> and if
a <slot> does not have assigned nodes, it treats its real child nodes as the
children.
nsINode::GetChildAt<TreeKind::Flat>(),
nsINode::GetChildCount<TreeKind::Flat>() and
nsINode::ComputeIndexOf<TreeKind::Flat() of a shadow host
treat the shadow root children as the children.
nsINode::GetParentNode<TreeKind::Flat>() of an assigned node of a <slot>
returns the <slot> element.
nsINode::GetParentNode<TreeKind::Flat>() of a child of a
ShadowRoot returns the host. So, ShadowRoot is ignored.
nsINode::GetParentNode<TreeKind::Flat>() of a ShadowRoot
returns nullptr.
The UA shadow trees are not ignored. So, we could say this treats the flattened tree as what you see.
When you call nsINode::GetParentNode<TreeKind::Flat>() on a child of a
<slot> which has some assigned nodes, it’ll return nullptr because the node
is not a part of the flattened tree.
TreeKind::FlatForSelection
Handle the flattened tree except the
UA shadow trees. The assigned nodes of <slot> elements are
treated as children of the <slot> element (if a <slot> does not have
assigned node, it’s treated as a normal element which may have some children).
The normal flattened tree does not contain non-flattened nodes. From Selection API point of view, they can be a container of a range boundary. Therefore, we need to treat them as a part of the flattened tree for selection especially when we retrieve ancestor nodes.
nsINode::GetChildAt<TreeKind::FlatForSelection>(),
nsINode::GetChildCount<TreeKind::FlatForSelection>() and
nsINode::ComputeIndexOf<TreeKind::FlatForSelection() of a <slot> which has
some assigned nodes treat the assigned nodes as the children of the <slot>.
nsINode::GetChildAt<TreeKind::FlatForSelection>(),
nsINode::GetChildCount<TreeKind::FlatForSelection>() and
nsINode::ComputeIndexOf<TreeKind::FlatForSelection() of a
shadow host treat the shadow root children as the children.
nsINode::GetParentNode<TreeKind::FlatForSelection>() of an assigned node of
a <slot> returns the <slot> element.
nsINode::GetParentNode<TreeKind::FlatForSelection>() of a child of a
ShadowRoot returns the host. So, ShadowRoot is ignored.
nsINode::GetParentNode<TreeKind::FlatForSelection>() of a
ShadowRoot returns nullptr.
nsINode::GetParentNode<TreeKind::FlatForSelection>() of a child of a
shadow host and not assigned to a <slot> returns the shadow
host.
nsINode::GetParentNode<TreeKind::FlatForSelection>() of a child of a <slot>
which has some assigned nodes returns the <slot>.
The UA shadow trees are ignored too. Similarly, when you
handle a <slot> element which has some assigned nodes, the APIs for
TreeKind::FlatForSelection check whether the <slot>’s ShadowRoot is a UA
one. If it’s so, the <slot> element is treated as a usual container element.
Iterating children of a node
The children of Shadow host element may not a part of
the flattened tree or assigned to a <slot> in the shadow.
Therefore, it may not be cheap to get the siblings of a child node in the
flattened tree. Therefore, nsINode::GetFlattenedTreeNextSibling() etc are not
available to make the developers realize the cost.
To handle multiple siblings, there is a useful template iterator class,
ChildIterBase<TreeKind>. There are alias names for each TreeKind. (To
lead the developers to this template class, there are the deleted nsINode
API declarations.)
|
|
|---|---|
|
|
|
N/A |
|
|
|
|
You can initialize them with a parent node.
ChildIterBase<aKind> iter(parentNode);
Then, if you need to iterate from a specific child rather than from the first child, you need to seek the child.
if (!iter.Seek(childNode)) {
return; // childNode is not a child of parentNode in the specified tree.
}
Then, you can iterate the remaining children:
for (nsIContent* sibling = iter.GetNextChild(); sibling;
sibling = iter.GetNextChild()) {
// Do something with sibling.
}
for (nsIContent* sibling = iter.GetPreviousChild(); sibling;
sibling = iter.GetPreviousChild()) {
// Do something with sibling.
}
Comparing points
nsContentUtils::ComparePoints and nsContentUtils::ComparePointsWithIndices()
are template methods whose template parameter is TreeKind.
If TreeKind is DOM, the points in different shadow trees are treated as
disconnected. Otherwise, the points are compared across the shadow DOM
boundaries.
Even if the TreeKind of the template method is ShadowIncludingDOM, the
points can be the points in the flattened tree. Similarly, even if the
TreeKind is FlatForSelection, the points can be the points in the DOM.
Using different TreeKind points are not recommended, but works in the most
cases. However, to prevent a bug in an edge case, you should convert the
points to TreeKind::DOM if you use TreeKind::ShadowIncludingDOM method and
convert the points to TreeKind::FlatForSelection or TreeKind::Flat when you
use the corresponding method.
(Probably, we should stop allowing different TreeKind range boundaries as
the parameters of nsContentUtils::ComparePoints.)
Non-flattened nodes are treated as disconnected in
TreeKind::Flat. However, they are treated as connected in
TreeKind::FlatForSelection. Unassigned nodes in
TreeKind::FlatForSelection are treated as if they were at end of the
corresponding shadow root. Similarly,
unused fallback content in
TreeKind::FlatForSelection are treated as if they were after the assigned
nodes of the <slot> element.
Convert a point in TreeKind::DOM or TreeKind::FlatForSelection to the other
RangeBoundaryBase supports TreeKind::DOM and TreeKind::FlatForSelection
and it has factory methods to create the instance in those kinds of tree.
Use AsRangeBoundaryInDOMTree() to get a point in TreeKind::DOM and use
AsRangeBoundaryInFlatTreeOrNonFlattenedNode() to get a point in
TreeKind::FlatForSelection. If the referring child node has different parent
in the trees, these methods compute the proper parent automatically.
When AsRangeBoundaryInFlatTreeOrNonFlattenedNode() converts a TreeKind::DOM
point, the container may be a non-flattened node. If the referring child is not
flattened by the parent shadow host or <slot>, this converts
to start or end of the corresponding ShadowRoot or the <slot>. (Only when
the boundary is start of the container, it converts to the start. Otherwise,
converts to the end because non-flattened nodes are treated as if they were
at end of the container.)
GetRangeBoundaryInFlatTree() returns a point in flattened tree whose
container is never a non-flattened node. If the given
point is in an unassigned node, it’s converted to the end
of the corresponding shadow root (i.e., after the last child of
the shadow root). Similarly, if the given point is in
unused fallback content, it’s converted to the end
of the <slot> element (i.e., after the last assigned node).
nsINode API list
|
|
|---|---|
|
|
|
N/A |
|
|
|
|
|
|
|---|---|
|
|
|
N/A |
|
|
|
|
|
|
|---|---|
|
|
|
N/A |
|
|
|
|
|
|
|---|---|
|
|
|
N/A |
|
|
|
|
|
|
|---|---|
|
|
|
N/A |
|
|
|
|
|
|
|---|---|
|
|
|
N/A |
|
N/A |
|
N/A |
|
|
|---|---|
|
|
|
N/A |
|
|
|
|
|
|
|---|---|
|
N/A (The template API returns |
|
|
|
|
|
|
nsIContent API list
|
|
|---|---|
|
N/A (The template API returns |
|
N/A (The template API returns |
|
|
|
|